Field note

How I use AI without making slop

I've always made things with whatever the newest tools were. At The Wall Street Journal that meant a camcorder, a laptop, and no crew. At Spire, it meant building a flagship campaign, The Promise of Space, with generative creative before most brand teams had tried it. At Eightfold, it meant a brand film made with Runway and Seedance.

So when I say I'm against slop, I don't mean I'm against AI. I mean something narrower, and I think more useful.

What slop actually is

Slop isn't AI. It's AI with nobody steering.

It's the homepage that could belong to any company in the category. It's the stock-perfect "customer" who never existed, the testimonial nobody said, the video that looks expensive and says nothing. It gets made because it was cheap and fast, not because it was right.

Buyers can tell. They may not name it, but they feel it, and what they feel is that nobody cared. That feeling is the opposite of what a brand is for.

My four rules

1. Say when it's generated. Anything made with generative AI gets labeled. On my site, those pieces carry a GenAI badge. Disclosure costs nothing, and it's the difference between a tool and a trick.
2. A person directs every frame. The story, the cut, and the final call stay human. The model is a crew, not a director.
3. Use it where a camera can't go. Orbit. Scale. A product that doesn't exist yet. The inside of a data center, or the future a company is building toward. That's where generative work earns its place.
4. Never fake reality. No generated customers, employees, events, or testimonials presented as real. If a real person or real footage exists, it wins.

Where AI does belong

Most of the value isn't in the showy parts. It's in the volume work nobody sees: turning hours of customer interviews into a searchable library of what people actually said, checking every claim against the approved messaging before it ships, and cutting one film into every size and channel in minutes instead of days.

That work used to eat a team's week. Now it frees them for the part that matters, which is deciding what to say and making it true.

The test

I spent nearly twenty years as a reporter, and I still use a reporter's test before anything ships: would I be comfortable if a journalist asked how this was made?

If the answer is yes, and I can explain every choice, the tool doesn't matter. If the answer is "I'd rather they didn't ask," it's slop, however good it looks.

The tools will keep getting better. The question stays the same: who's steering?